The Agent Skills Directory

[Security Architecture]: The skill documents a 'Gatekeeper' architecture designed to mediate access between AI agents and external APIs. This design enforces security boundaries by requiring explicit user approval for any side-effecting actions via a structured ApprovalQueue.
[Capability-Based Design]: It encourages implementing fine-grained resource access (e.g., access to one specific document rather than a whole account), which aligns with the principle of least privilege.
[Observer Verification]: The implementation guidance includes mandatory methods for verifying collaborator access (getVerifier, addObserver), ensuring that sensitive data is not leaked when a workspace is shared.
[Secure Coding Practices]: The provided code skeleton demonstrates standard security measures, such as constant-time equality checks (timingSafeEqual) for nonces to prevent timing attacks and the use of the allow_irrevocable_stub_storage flag for persistent RPC connections.
[Credential Management]: The skeleton correctly guides developers to store sensitive tokens within Durable Object storage rather than in less secure contexts, and implements self-destruct alarms for uncompleted authentication flows.

write-gatekeeper